Um estudo sobre o erro de estimação de movimento translacional em imagens por algoritmos baseados na correlação de fase

Abstract

Image Registration is the term used to describe the process of aligning two images. This process can be described as the estimation of the relative movement between objects (pixels) of an image in relation to another, containing similar contents. Algorithms based on Phase Correlation are of special interest because they are assumed to be fast, which makes them suitable for realtime processing. The works found in the literature on the behavior of these algorithms focus on the influence of the aliasing generated by rotation on the estimation/record error. Recently, a study has pointed to innovations (objects present in only one of the images to be registered) as equally or even more important, as is already assumed in other registration approaches. This work proposes a study on the influence of innovations on the registration error, by the Phase Correlation method, for translational movements. As a result, this study points out possible directions for an investigation about the reliability of the motion estimation by this method. [resumo fornecido pelo autor]